基于观察学习的机场噪声异构集成预测模型

发布时间：2018-12-16 12:32

【摘要】：随着我国综合国力的不断提升,民航运输业取得了非常瞩目的发展成果。然而日益严重的机场噪声污染问题也伴随而来,有效控制机场周边噪声污染已成为民航从业人员当前必须重点解决的一个问题。机场噪声预测是进行机场噪声评估和噪声防治工作的重要前提,因此,构建科学合理全面的机场噪声预测模型意义非凡。本文详尽地研究了已有的基于机器学习的机场噪声预测方法,其中关联分析预测多采用单一学习器,预测精度不高且泛化能力差。为此,本文提出了一种机场噪声关联分析集成预测方法,该方法综合考虑机场噪声的主要影响因素,结合集成学习的思想,利用空间拟合算法和BP神经网络算法构建多个基学习器,然后采用观察学习算法将多个基学习器集成。通过多个不同学习器的集成,能够有效提升关联分析预测的准确性,同时多个异构学习器的集成保证了预测方法的泛化能力。基于卡尔曼滤波的思想,通过在时间序列预测结果上的改进,本文提出一种基于卡尔曼滤波优化的机场噪声时间序列预测方法。该方法设计以噪声统计值构造时序的方案,利用支持向量回归机训练预测学习器,然后采用卡尔曼滤波对预测结果进行去干扰优化。合理的时序构造方案和预测结果的去干扰优化,使得本文的方法较之前人的时间序列预测,准确率有显著地提升。最后,针对机场实际噪声预测稳定准确可靠的需求,本文提出一种基于观察学习的机场噪声异构集成预测模型。该模型通过一致化数据集,利用反向观察学习将关联分析预测和时间序列预测这两种异构的预测方法合理地集成起来。该模型较之单一方法的噪声预测,具有更高的预测精度和更强的泛化能力,适用于我国大部分机场的实际应用状况。
[Abstract]:With the continuous improvement of China's comprehensive national strength, the civil aviation transportation industry has made remarkable achievements. However, the increasingly serious problem of airport noise pollution has been accompanied, effective control of noise pollution around the airport has become a problem that civil aviation practitioners must focus on at present. Airport noise prediction is an important prerequisite for airport noise assessment and noise prevention, so it is of great significance to build a scientific, reasonable and comprehensive airport noise prediction model. In this paper, the existing prediction methods of airport noise based on machine learning are studied in detail. Among them, only one learner is used for correlation analysis prediction, and the prediction accuracy is not high and the generalization ability is poor. In this paper, an integrated prediction method for airport noise correlation analysis is proposed. The method considers the main influencing factors of airport noise and combines the idea of integrated learning. The spatial fitting algorithm and the BP neural network algorithm are used to construct a number of basic learning devices, and then the observation learning algorithm is used to integrate the multiple basic learning devices. The integration of multiple different learning devices can effectively improve the accuracy of correlation analysis and prediction, and the integration of multiple heterogeneous learning devices ensures the generalization ability of prediction methods. Based on the idea of Kalman filter and the improvement of time series prediction results, this paper presents a time series prediction method for airport noise based on Kalman filter optimization. In this method, a scheme of constructing time series with noise statistics is designed. Support vector regression machine is used to train predictor, and Kalman filter is used to optimize the prediction results. Due to the reasonable time series construction scheme and the optimization of the prediction results, the accuracy of this method is significantly improved compared with the previous time series prediction. Finally, a heterogeneous integrated prediction model of airport noise based on observation learning is proposed to meet the demand of stability, accuracy and reliability of airport noise prediction. In this model, two heterogeneous prediction methods, association analysis prediction and time series prediction, are reasonably integrated by using reverse observation learning through consistent data sets. Compared with the single method, this model has higher prediction accuracy and stronger generalization ability, which is suitable for the practical application of most airports in China.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：V351;TP18

【相似文献】